DeepSeek releases R1, a new open-weights reasoning model that ranks #2 globally, signaling intensifying competition in AI development.

What happened

DeepSeek introduced R1, an open-weights reasoning model (AI software available for anyone to download and modify) that has been ranked as the #2 open-weights reasoning model. The model uses 32T–33T tokens of training data and achieves 1M token context length (the amount of text it can process at once), up from 128K in V3.2.

Why it matters

Open-weights models like R1 are freely available, unlike proprietary systems from leading labs, which means any developer or company can build on top of it. A top-ranking open-weights model increases options for businesses and researchers who want to avoid dependency on a single commercial vendor.

What to watch

The #2 ranking among open-weights reasoning models—meaning only one open-weights model ranks higher—reflects how quickly the competitive landscape is shifting in AI model development.